pool_blocks crashes on single-trial blocks

[lschwetlick]

Minimal repro:

bla2 = np.array([
    [100, 50, 50],
    [101, 51, 51],
    [200, 60, 60],
    [201, 61, 61],
    [300, 70, 70],
    [301, 71, 71],
    [400, 80, 80]
])

# Crashes with:
# TypeError: cannot unpack non-iterable numpy.float64 object
pooled = ps.pool_blocks(bla2, max_tol=50, max_gap=50, max_length=10)

Two things:

• max_length refers to cumulative trial count (not number of rows), which isn’t super clear in the doc.
• If pooling results in just one trial (because one line already exceeds the max_length), the code crashes — probably needs to reshape the block before summing.

we should either error gracefully or reshape before.

[lschwetlick]

Taking votes for preferred behavior before I fix it

[pberkes]

I think every block should have at least one line, so in the example each line is a separate block because in all lines ntrials > max_length?

[guillermoaguilar]

I don't really understand why it crashes, but I also don't get what would be the correct output in the above case. Should it pool together 100 with 101, 200 with 201 and 300 with 301? That's what I'd expect, so the desired output only 4 stimulus levels... is that so?

Two things:

• max_length refers to cumulative trial count (not number of rows), which isn’t super clear in the doc.

it's also not clear to me what the arguments mean. Intuitively, max_length is the maximum number of stimulus levels desired/allowed, right?

[pberkes]

If I read the docstring and code correctly, the output should be equal to the input, because each row already has a number of trial larger than max_length.

Can anyone try in Matlab and check what the expected output is in the original implementation?

[guillermoaguilar]

I just tried it in MATLAB, it pools in such a way that the stimulus values are at 100, 200, 300 and 400.

[guillermoaguilar]

no sorry, the MATLAB version does not pool this case. In the visualization the dots are just overlapping, it has actually all 7 dots.
So the MATLAB version does not crash, but also it does not pool.

[pberkes]

then do we agree that the output should be equal to the input (because n_trials > max_length for each row)? then the problem is that the current code freaks out if there is only on row to pool, it should be an easy fix

[guillermoaguilar]

then do we agree that the output should be equal to the input (because n_trials > max_length for each row)? then the problem is that the current code freaks out if there is only on row to pool, it should be an easy fix

yes

[guillermoaguilar]

then do we agree that the output should be equal to the input (because n_trials > max_length for each row)? then the problem is that the current code freaks out if there is only on row to pool, it should be an easy fix

could you take care of it @pberkes ? that would be lovely, I'm super busy atm :/